To determine whether activation of CGRP^PBN neurons is sufficient to induce both the behavioral and physiological correlates of the unconditioned response in addition to fostering associative fear learning (Han et al., 2015), we bilaterally injected an adeno-associated virus carrying Cre-dependent channelrhodopsin (AAV1-DIO-ChR2:YFP) and implanted fiber-optic cannulae over the PBN of Calca^Cre/+ mice, while control mice received AAV1-DIO-YFP (Figure 1A; Figure 1—figure supplement 1A). Repeated high-frequency (30 Hz) activation of CGRP^PBN neurons induced profound freezing behavior (Figure 1B, Figure 1—figure supplement 1B–C and Figure 1—video 1), as indicated by rigid, uninterrupted immobility. In addition to eliciting robust freezing behavior, we confirmed that pairing photostimulation with an auditory CS rapidly induced conditioned freezing responses to the CS (Figure 1C; Han et al., 2015). To test whether CGRP^PBN neurons can recapitulate physiological aspects of the unconditioned response, we photostimulated the neurons while monitoring heart rate with a pulse oximeter (Figure 1D). Interestingly, while modest activation (15 Hz, subthreshold for eliciting freezing behavior) resulted in moderate tachycardia (Figure 1—figure supplement 1E), high-frequency activation (30 Hz) led to profound bradycardia and decreased respiration followed by dramatic post-stimulation rebound tachycardia and mild hyperventilation (Figure 1E–F, respiration measured in plethysmography chamber); it also produced vasoconstriction (Figure 1—figure supplement 1H; Vianna and Carrive, 2005). Hence, CGRP^PBN neurons are capable of exerting opposing effects on autonomic physiology depending on their activation frequency. Comparing the latencies of somatic vs autonomic responses to 30 Hz photostimulation, we found that freezing behavior is rapidly initiated (median 0.42 s), while bradycardia takes longer to develop (median 22.15 s, Figure 1—figure supplement 1C–D), suggesting that freezing behavior does not emerge simply as a consequence of adverse autonomic effects. To test whether CGRP^PBN neurons can also elicit behavioral alterations associated with late-phase responses to threat exposure, we subjected mice to an elevated-plus-maze test (Martin, 1961; Pellow et al., 1985) while activating CGRP^PBN neurons; this treatment attenuated open-arm exploration consistent with an anxiogenic effect (Figure 1G).

Figure 1 with 2 supplements see all

Download asset Open asset

To map the forebrain connections from CGRP^PBN neurons that underlie their wide physiological and behavioral repertoire, we sectioned the forebrain of mice expressing a fluorescent tracer (AAV1-DIO-YFP) in CGRP^PBN neurons and identified axon terminals in various downstream sites (Figure 1H). Comparing individual targets to cumulative projection intensity, we found major projections to the central amygdala (CeA,~40%, primarily targeting the capsular sub-nucleus), substantia innominata (SI,~20%), and oval sub-nucleus of the bed nucleus of stria terminalis (ovBNST,~15%), with weaker projections to the parasubthalamic nucleus, thalamus and visceral insular cortex; PSTN and VPMpc,~10% each, IC,~5% (Figure 1I, for abbreviations see figure supplement 2). With the exception of the IC, CGRP^PBN neurons also target the contralateral hemisphere for all of their downstream partners, markedly to the contralateral PSTN and VPMpc, with ~75% and 50% of the ipsilateral projection intensity, respectively (Figure 1—figure supplement 1I–J). To confirm that downstream neurons receive monosynaptic excitation from CGRP^PBN neurons and also to compare synaptic strength across targets, we expressed channelrhodopsin (ChR2) in CGRP^PBN neurons and photostimulated terminals in downstream regions while recording from putative postsynaptic neurons in a slice preparation (Figure 1J). Interestingly, we found that while all the major downstream targets were recipients of reliable excitatory input from CGRP^PBN neurons (Figure 1J, IC not tested), the VPMpc, while not receiving the strongest input based on fiber density, exhibited significantly greater excitation from terminal activation than any other recording site (Figure 1K).

The heterogeneity of behavioral and physiological outcomes elicited by activation of CGRP^PBN neurons raises questions about the underlying circuit organization responsible for their generation. We envisioned several potential circuit structures underlying CGRP^PBN neuron connectivity to the forebrain: while distributed, one-to-all connectivity involving extensive collateralization from each CGRP neuron to every target structure would be well suited for simultaneous, parallel activation of diverse regions, a one-to-one, segregated organization would better support separable generation of distinct functions via activation of designated partners. To reveal the structure underlying CGRP^PBN-neuron connectivity to the forebrain, we devised a method to selectively isolate subsets of CGRP^PBN neurons as defined by their target-projecting behavior. By injecting AAV expressing retrogradely transported Flp-recombinase (rAAV2-retro-Flp) into a downstream site and a fluorescent tracer requiring both Cre and Flp for expression (Fenno et al., 2014) (AAV-Cre_on-Flp_on-YFP; Target +), or that is turned on by Cre but off by Flp (Fenno et al., 2014) (AAV-Cre_on-Flp_off-YFP; Target –) into the PBN of Calca^Cre/+ mice (Figure 2A), we were able to isolate fluorescent expression to neuronal subpopulations defined by whether or not they targeted a region of interest (Figure 2B, Figure 2—figure supplement 1A). Normalizing the resulting projection intensity in each downstream region under each condition to the maximal signal given by transducing all CGRP^PBN neurons, we determined the proportion of terminal density in each downstream partner supplied by target-projecting CGRP^PBN neurons for the VPMpc, PSTN, CeA, and ovBNST (Figure 2C). This analysis revealed that CeA-projectors contributed substantially to PSTN, SI, VPMpc and ovBNST projections, but not IC. VPMpc-projectors, interestingly, while also projecting to the CeA, contributed more substantially to the SI and IC, while PSTN-projectors had limited secondary output to the CeA and SI, and ovBNST-projectors had only a weak secondary projection to the CeA (Figure 2C), shown schematically in Figure 2E. Quantifying the number and location of the different projecting subpopulations within the PBN revealed that neurons projecting to the CeA made up the largest proportion of CGRP^PBN neurons residing within the external lateral PBN, while neurons projecting to VPMpc accounted for most of the CGRP^PBN neurons residing in the medial and waist regions; neurons projecting to ovBNST, the smallest group, were restricted to the external lateral PBN (Figure 2—figure supplement 1B–F). Comparing projection distributions for the Target + or Target - expression conditions, we found that regardless of the downstream target used to drive expression, the CeA was the primary downstream partner in terms of projection intensity (Figure 2D). Excluding CeA-projecting CGRP^PBN neurons flattened the distribution, with the ovBNST narrowly making up the largest projection contribution. As a summary statistic to directly compare the collateralization tendencies across subpopulations, we calculated a collateralization coefficient defined as the difference between projection strength for each downstream partner in the Target + and Target - conditions, for each target, where a value of 50% corresponds to half of the signal in the area of interest being supplied by target-site projectors (Figure 2F, Figure 2—figure supplement 1G). Looking at the distribution of these coefficients across secondary downstream partners for each target site, we found that VPMpc projectors had the greatest tendency to collateralize, while ovBNST projectors collateralized primarily to the CeA (Figure 2G, Figure 2—figure supplement 1G). In summary, there is extensive collateralization by CGRP^PBN neurons with no one-to-one projections; rather, CGRP^PBN neurons tend to distribute their projections among large groups of downstream targets, composing a one-to-many distributed projection arrangement (Figure 2E).

Figure 2 with 1 supplement see all

Download asset Open asset

To assess the contribution of activating CGRP^PBN projections to individual brain regions in eliciting behavioral and physiological processes associated with unconditioned responses to aversive stimuli, we used ChR2 to stimulate terminals within specific target regions (Figure 3—figure supplement 1A–C), for fiber placement summary; fibers targeting the caudal and rostral CeA were placed in the caudal or rostral third of the CeA (caudal to −1.4 AP and rostral to −1.0 AP, respectively). Because of the high degree of collateralization, it is possible that stimulating one region will result in antidromic activation and neurotransmitter release in all areas with shared innervation. If that occurred, then stimulating in one area that shares strong co-innervation with another should yield similar phenotypic outcomes. Surprisingly, given the broad collateralization of CGRP^PBN neurons, that was not the case. Only photostimulating terminals in the VPMpc or PSTN led to reliable initiation of freezing behavior (Figure 3A–B,~40% time-spent freezing), while photostimulating the caudal CeA (cCeA), SI, or ovBNST had more subtle effects (~25% time-spent freezing, Figure 3C–F), and stimulating the rostral CeA (rCeA) actually led to a non-significant increase in locomotion (Figure 3—figure supplement 2B; for cross-area mean freezing response comparisons see Figure 3—figure supplement 2A). Notably, activating no individual projection was able to match CGRP^PBN cell-body activation in generating robust freezing behavior (Figure 3—figure supplement 2A,C). Importantly, while freezing behavior is an unconditioned response to predator incursion (Roelofs, 2017), when elicited by noxious stimulation it is instead a learned response to contextual cues because adaptive responses to ongoing noxious stimulation are always to flee or withdraw (Fanselow, 1982; Landeira-Fernandez et al., 2006). To examine whether the freezing behavior we observed was directly elicited by photostimulation or was instead driven by processes secondary to contextual conditioning, we looked at the temporal structure of the freezing responses to light onset and offset. We found that photostimulation led to short-latency freezing bout initiation (<5 s after stimulation onset) for most terminal stimulation groups except the rCeA, which instead elicited short-latency freezing bouts after stimulation offset (~2.4 s) (Figure 3—figure supplement 2C–D). Freezing-bout initiation occurred with lower latencies than control animals during the 20 s post-stimulation epoch in all stimulation groups except the VPMpc and cCeA (Figure 3—figure supplement 2C–D). When taken together with the observation that freezing behavior occurs with greater frequency during the stimulation epoch than post-stimulation epoch for all fiber-placement groups except those in the CeA and ovBNST (Figure 3—figure supplement 2E), these findings suggest that stimulation of most CGRP^PBN neuron projections simultaneously elicits a direct effect on freezing behavior while also generating aversive properties that promote transient contextual freezing. Interestingly, in the case of the rCeA, the direct effect on freezing is absent, while the contextual memory effects on freezing are instead the primary effect (Figure 3—figure supplement 2C–E).

Figure 3 with 3 supplements see all

Download asset Open asset

By measuring the effect of photostimulating different terminal fields on multiple physiological measures, we found that activating the PSTN, rCeA, SI or ovBNST led to tachycardia, while activating the VPMpc or cCeA had no effect (Figure 3F–L). In addition to eliciting tachycardia, photostimulating terminals in the PSTN, rCeA or SI caused vasoconstriction (Figure 3S–X), while activating only the rCeA, SI, or ovBNST elicited hyperventilation (Figure 3M–R). Lower frequency stimulation (15 Hz) led to similar, less robust physiological effects across regions (Figure 3—figure supplement 3A–F), while light delivery alone in control animals had no effect on any of these measures (Figure 3—figure supplement 3G–L). Compellingly, the most co-innervated downstream regions – the VPMpc and SI, CeA and ovBNST, and PSTN and CeA, each had distinct effects on physiology and behavior, with some (VPMpc, PSTN) preferentially inducing freezing behavior, and others (SI, CeA, ovBNST) robustly eliciting autonomic responses, suggesting that terminal stimulation does not produce robust antidromic activation that homogeneously activates all co-innervated regions. In support of this conclusion, we observed that photostimulation of terminals in each downstream target did not generate antidromic activation of CGRP^PBN cell bodies sufficiently to induce Fos expression in the PBN (Figure 3—figure supplement 2G). Taken together, these behavioral and physiological data suggest that the projections to thalamic (VPMpc) and hypothalamic (PSTN) targets elicit freezing behavior the best, while activating extended amygdalar structures (rCeA, SI, ovBNST) elicits sympathetic autonomic responses, implying a specialization in function across downstream partners.

To measure alterations in anxiety state, potentially indicative of enhanced arousal or vigilance in response to threats (Martin, 1961; Mestanik et al., 2015), we photostimulated terminals in downstream targets while mice explored an elevated-plus maze (Figure 4A). Only photostimulation of terminals in the ovBNST significantly reduced open-arm exploration, consistent with an anxiogenic effect, while photostimulating terminals in the rCeA paradoxically increased open-arm exploration (Figure 4B–C).

Figure 4 with 1 supplement see all

Download asset Open asset

To further interrogate the affective state generated by activation of each downstream partner we utilized a real-time, place-preference (RTPP) assay to assess whether mice would choose to seek out or avoid terminal photostimulation (Figure 4D). Mice with photostimulation of either CGRP^PBN somata or their terminals in the VPMpc, PSTN, rCeA, or SI robustly avoided photostimulation (Figure 4E–K, Figure 4—figure supplement 1E–K), whereas mice with photostimulation of terminals in the cCeA or ovBNST had no preference relative to control animals, which spent equal time in the three compartments. Considering aversive valence in combination with the observation that photostimulation of terminals in the rCeA robustly potentiated escape attempts during exposure to noxious heat (Figure 4—figure supplement 1A–C) without affecting spinal analgesia (Figure 4—figure supplement 1D), implies that activating the rCeA may not be anxiolytic per se, but shift behavior toward active coping strategies during threatening situations (D’amour and Smith, 1941; Espejo and Mir, 1993).

While we had observed that stimulation of multiple individual projections was able to transiently generate contextual freezing, we were interested in distinguishing between intrinsic effects of stimulation on freezing behavior versus secondary effects on associative learning. To accomplish this, we subjected mice to an associative fear-learning paradigm where an auditory conditioning stimulus (CS) precedes and co-terminates with terminal photostimulation as an unconditioned stimulus (US) to assess the ability of activating each individual projection target to generate a fear memory, revealed by testing for conditioned responses to the CS in a novel environment (Figure 5A). Photostimulation of CGRP^PBN-neuron terminals in the VPMpc, PSTN, or SI resulted in significant freezing to the auditory CS after 6 CS-US pairings (Figure 5B–G), with only activation of terminals in the VPMpc or SI generating a significant association as indicated by area under the curve exceeding that of control animals (Figure 5H) and robust conditioned freezing to the CS in a novel context 24 hr following conditioning (Figure 5B–G). While photostimulation of CGRP^PBN neuron terminals in either the SI or VPMpc was sufficient to drive associative fear learning, the association formed is weaker than that driven by photostimulating CGRP^PBN neuron cell bodies (Figure 5H), suggesting they play complementary roles.

Figure 5

Download asset Open asset

Activation of no single projection from CGRP^PBN neurons was sufficient to elicit profound freezing behavior or bradycardia; therefore, we devised a method to simultaneously activate multiple terminal fields by implanting three fiber-optic cannulae in a single hemisphere over multiple areas of interest to determine the threshold of downstream activity necessary to elicit these phenotypes (Figure 6A). We placed one cannula over the SI, one over the cCeA and one over the VPMpc. Then, we determined the strength of freezing responses capable of being generated by each individual projection field by varying the light power. Maintaining stimulation frequency at 30 Hz and increasing laser power from 10 to 40 mW, we found that activation of CGRP^PBN neuron terminals in the cCeA or VPMpc led to a gradual increase in freezing but activating terminals in the SI was maximal at 10 mW (Figure 6B). Combining photostimulation of terminals in the VPMpc and SI (10 mW each) led to rapid entrainment of freezing behavior to an auditory CS, and the resulting association strength, although not significantly greater than either projection individually generated, was no longer significantly weaker than that generated by the entire population even though our dual-stimulation arrangement was unilateral and all other groups were bilateral (Figure 6C).

Figure 6 with 3 supplements see all

Download asset Open asset

We then combined activation of multiple projection fields using 20 mW power to determine which combination of CGRP^PBN neuron projections could elicit profound freezing behavior. Activating the cCeA and SI projection fields resulted in moderate freezing behavior that did not appear to be additive (cCeA 23.2 ± 0.5% freezing; SI 26.9 ± 2.5; Combined 30.4 ± 1.3 (mean ± sem), Figure 6D; Figure 6—video 1), while simultaneous activation of terminals in the VPMpc and cCeA elicited robust freezing behavior exceeding that produced individually (VPMpc 37.1 ± 2.3% freezing; cCeA 23.2 ± 0.5; Combined 68.9 ± 3.6 (mean ± sem), Figure 6B and E–F; Figure 6—video 2), comparable to freezing behavior elicited by activating all CGRP^PBN neurons bilaterally (93.0 ± 2.9% freezing (mean ± sem), Figure 1B; Figure 6—video 1). These phenotypes were enhanced by driving photostimulation with a red light-activated opsin (Yizhar et al., 2011) (VPMpc+cCeA 94.5 ± 3.0% freezing (mean ± sem), Figure 6—figure supplement 1A–I), suggesting that a combination of light-spread and faithfulness of activation underlies reliable freezing generation. Importantly, simultaneous photostimulation of terminal fields did not dramatically induce Fos in CGRP^PBN neurons (<10% compared to 80% for soma activation sufficient to generate freezing behavior) (Figure 6—figure supplement 1J–L). We also tested whether simultaneous photostimulation of terminals in the cCeA and VPMpc would affect autonomic physiology by measuring heart rate using a pulse oximeter. While activating neither projection alone affected heart rate (Figure 3), simultaneous photostimulation robustly elicited bradycardia, which consistently took longer to develop than freezing responses (Figure 6G, Figure 6—figure supplement 1A). These results imply that when combinations of projections from CGRP^PBN neurons are activated simultaneously their combined output is able to generate phenotypes beyond their individual ability, suggesting a mechanism by which defensive responses can be tuned depending on whether downstream partners are already in an ‘up’ or ‘down’ state as determined by broader sensory context when they receive input from CGRP^PBN neurons.

While previous studies that permanently silenced CGRP^PBN neurons demonstrated that their activity contributes to conditioned-fear responses (Han et al., 2015), we asked whether photoinhibition restricted to the peri-foot shock period during conditioning would be sufficient to attenuate conditioned responses to the CS, as post-shock recurrent activity, stress-induced activation, or recall-driven reactivation could also potentially affect association formation, memory consolidation, or recall. Using AAV-mediated expression of a red-light activated chloride pump (Chuong et al., 2014) (JAWS) to inhibit CGRP^PBN neurons during 0.5-mA foot-shock delivery (Figure 7A, Figure 7—figure supplement 1A), we found that selective inhibition of CGRP^PBN neurons during the foot shock significantly attenuated both conditioned responses during training and in a CS-probe trial 24 hr later, while also reducing freezing behavior conditioned to the training context (Figure 7B). These findings affirm that the signal relayed by CGRP^PBN neurons to downstream partners during the foot shock directly contributes to associative memory formation.

Figure 7 with 1 supplement see all

Download asset Open asset

To determine whether individual projections contribute to associative fear learning, we used JAWS to inhibit CGRP^PBN-neuron terminals in the VPMpc, CeA, or SI during the foot shock (Figure 7C). We first confirmed that JAWS-mediated inhibition of CGRP^PBN-neuron terminals significantly reduced EPSC frequency in post-synaptic neurons (Figure 7D; Mahn et al., 2016). Inhibiting synaptic release during the foot shock at CGRP^PBN-neuron terminals in the VPMpc or SI, but not CeA, significantly attenuated both memory formation (Figure 7E–G) and association strength (Figure 7H), without affecting contextual-fear learning. While inhibiting CGRP^PBN neurons non-significantly reduced foot shock-induced locomotion (Figure 7I), no individual projection tested was necessary for this response. In addition, transiently inhibiting either CGRP^PBN cell bodies or their individual projections did not significantly affect behavioral responses to noxious heat (Espejo and Mir, 1993; Figure 7—figure supplement 1B–E), nor did it lead to a place preference in a RTPP paradigm (Figure 7—figure supplement 1F–I), suggesting that basal activity of CGRP^PBN neurons is insufficient for their inhibition to generate a salient shift in affective state. Taken together, these data reveal an unexpected role for the SI and VPMpc, two regions respectively implicated in arousal (Kaur et al., 2017; Mogenson et al., 1985) and taste processing (Liu and Fontanini, 2015), in contributing to an affective pain signal that drives associative fear learning.

Previous studies silencing CGRP^PBN neurons implicated them in contributing to both affective responses to somatic pain, including nocifensive behavior, post-shock freezing behavior (Han et al., 2015), and illness-induced increases in anxiety state (Campos et al., 2017). We found that photostimulation of CGRP^PBN neurons, in addition to driving profound freezing behavior, can also generate either tachycardia or parasympathetic responses depending on stimulation frequency, and elicit anxiety-like behavior. These findings collectively suggest that activation of CGRP^PBN neurons during somatic pain has the potential to contribute to many aspects of the unconditioned response cascade, from shock-induced locomotion to post-shock freezing behavior, autonomic responses including simultaneous enhancement of parasympathetic and sympathetic outflow, and post-insult anxiogenesis. A complication of this arrangement is that neither freezing behavior (Blanchard and Blanchard, 1969), parasympathetic responses (Iwata and LeDoux, 1988), nor anxiety occur during the shock. Hence, the role played by CGRP^PBN neurons in these phenotypes would necessarily result from recurrent reactivation, rather than a direct ascending signal.

By selectively activating CGRP^PBN-neuron terminals in their various downstream targets, we distinguished the potential of individual downstream partners to contribute to distinct components of the behavioral and physiological alterations that comprise the unconditioned- response cascade. We found that with the exception of the rCeA, all projections generated some amount of freezing behavior, with the most robust responses elicited by the PSTN and VPMpc, two projections that were overlooked in previous work. We also found a marked disparity in function across the CeA, with activation of terminals in the cCeA eliciting only mild freezing behavior, while activating the rCeA had no effect on freezing behavior during photostimulation but did produce robust sympathetic responses, brief contextual freezing following stimulation offset, avoidance, and nocifensive behaviors on a hot plate, all phenotypes reminiscent of responses to noxious stimulation. In general, our results suggest that CGRP^PBN-neuron connections to extended amygdalar structures (i.e. the CeA, SI, and ovBNST) influence freezing behavior, affective processing including negative valence and anxiety state, and physiological responses, while thalamic and hypothalamic connections transmit a negative-valence signal and elicit freezing behavior. These results are supported by the fact that, in rats, extended amygdalar structures are richly interconnected with hindbrain nuclei controlling autonomic outflow (Dong and Swanson, 2004; Rizvi et al., 1991; Veening et al., 1984), while the VPMpc is not (Cechetto and Saper, 1987). Our findings complement recent work that distinguished between PBN populations that target the extended amygdala and hypothalamus/periaqueductal grey and differentially drive affective and nocifensive responses, respectively (Chiang et al., 2020). However, by distinguishing between conditioned freezing responses and avoidance behavior, we were able to specifically implicate two novel targets, the VPMpc and SI, in generating associative fear learning, whereas Chiang et al., 2020 studies were limited to learned valence and only tested the CeA and ovBNST. We found that the PSTN, SI, and VPMpc also relay a negative valence signal. Hence, populations contributing to the negative valence of noxious or aversive stimuli may not necessarily contribute to associative fear learning, which has important implications for understanding the neural underpinnings of affective disorders such as PTSD.

Our collateral-tracing experiments revealed that, in contrast to the distinct phenotypes generated by terminal photostimulation, CGRP^PBN neurons form a broadly distributed network with their downstream partners in which no forebrain target receives solitary innervation. There was some bias in the connectivity groupings, with neurons projecting to the CeA tending to also strongly innervate the PSTN, neurons projecting to the VPMpc also innervating the SI and IC and avoiding the ovBNST, and neurons projecting to the ovBNST also targeting the CeA. Our findings are broadly in agreement with previous experiments that delineated sub-region-specific output and collateralization in rats (Sarhan et al.) and mice (Chiang et al., 2020), neither of which, however, reported connections or collateralization with the VPMpc, suggesting that cell-type specific expression more efficiently reveals this connection. Of interest, Sarhan et al., 2005 beautifully outlined rostral and caudal capsular CeA branching patterns across all extended amygdalar structures using single-axonal reconstructions of PBN neurons, finding collateralization between the rCeA and lateral hypothalamus (PSTN), and cCeA and ventral BNST. Taken as a whole, the distributed, collateralization organization of CGRP^PBN neurons may be important for generating highly coordinated actions and associations by simultaneously driving activity in downstream sites that have related or complementary functions. An example in support of this arrangement is that stimulation of terminals in the SI and VPMpc generated disparate effects on physiology, but collaboratively supported associative fear learning.

While activating some individual terminal fields from CGRP^PBN neurons in different downstream sites recapitulated – in a scaled-down fashion – most of the phenotypes driven by photostimulating the cell bodies, we found that profound freezing behavior and bradycardia were not produced by stimulation of any individual projection, suggesting they instead arise from additive interactions between downstream structures and their respective circuits. We tested this hypothesis by simultaneously activating terminals in the VPMpc and cCeA, two targets that generated reliable freezing behavior, and observed not only a robust potentiation of the freezing behavior but also profound bradycardia. Interestingly, neither of these populations generated autonomic responses when activated individually. One possible arrangement that explains this phenotype is that their concurrent activation gates activity in secondary structures that drive parasympathetic responses.

An important consideration in implicating individual downstream partners in generating distinct aspects of behavioral and physiological response is the inherent limitation of terminal photostimulation. It is difficult, if not impossible, to ensure that antidromic activity does not activate secondary targets, an especially important possibility given the broad collateralization of CGRP^PBN neurons. However, secondary techniques aimed at accounting for this situation also have their shortcomings: axons may bifurcate near the sites of interest rather than at the cell body, hence silencing cell bodies may not prevent antidromic activation. Moreover, since many of the forebrain structures contributing to threat processing are interconnected, silencing other portions of the downstream circuit to attempt to isolate the effect of the target of interest on the measured phenotype may affect phenotype generation if the populations are interconnected. We argue that the very fact that terminal stimulation in different downstream targets generates distinct phenotypes supports the fact that at minimum, preferential activation of the site of interest is occurring. If photostimulation of terminals was efficiently activating cell bodies within the PBN then the same phenotypes should be observed regardless of fiber location. Perhaps, most compelling is that CeA-projecting CGRP^PBN neurons make up the bulk of the population yet photostimulation of terminals in the CeA does not efficiently produce either freezing behavior or anxiety, two of the distinct phenotypes produced by activating other downstream targets that receive collateral innervation with the CeA.

Another necessary caveat of using artificial stimulation to probe the intrinsic functionality of different projection partners is that variation in transport rate and axon length across downstream partners could profoundly influence ChR2 levels at terminals and thus terminal stimulation efficacy, which would in turn confound the observed differences in function and connection strength across the circuit. We attempted to account for this by allowing 4 weeks after virus injection for ChR2 expression and transport before beginning experiments, having observed that 2–3 weeks was sufficient to observe robust labeling in the most distant terminal fields. While our electrophysiological measures revealed the greatest synaptic strength in CGRP^PBN neuron connections to one of their more proximal downstream partners, the VPMpc, the fact that the PSTN, which is equally close to the PBN but has a connection strength that is similar to what we observed with all other targets suggests that ChR2 transport cannot be the primary reason for the strength of the connection to the VPMpc. Moreover, we did not see a declination of synaptic strength across the proximal-distal axis – synapses in the ovBNST, the most distal target, exhibited similar synaptic strength to those 2 mm more proximal, in the PSTN. What then underlies the large differences in synaptic strength we observed across downstream targets that was independent of projection strength as measured by terminal labeling density? Differences in release probability, convergence (the VPMpc is a much smaller structure than the CeA, for example, hence each neuron might receive more contacts), postsynaptic receptor number or dendrite structure could all contribute to the observed differences in synaptic strength. One particularly interesting possibility is that the VPMpc, which does not express receptors for CGRP, may primarily rely on glutamatergic input from the PBN and hence exhibit large-amplitude EPSCs in response to CGRP^PBN neuron terminal photostimulation, while extended amygdala targets may rely more on CGRP release for activation (Shinohara et al., 2017; Okutsu et al., 2017), which our experiment would not have been able to reveal. A more detailed analysis of the electrophysiological properties of CGRP^PBN neuron to forebrain connections is necessary to understand the temporal and chemical underpinnings that help give rise to the different functions of each downstream connection.

Associative learning is a highly tractable and informative process because it reliably depends on the salience of the CS and US, and the innate associability of these stimuli (Garcia et al., 1968; Sigmundi et al., 1980). The interplay of these factors on the association is indicated by the learning rate and asymptote – the maximal conditioned response for a particular CS-US pair (Rescorla, 1972; Sigmundi et al., 1980). Here, we maintained a constant CS and varied the US by activating specific projections from CGRP^PBN neurons to condition predictive freezing to an auditory CS, or by silencing either CGRP^PBN neurons or individual projections during foot-shock delivery. Activation of CGRP^PBN neurons elicited the most robust association, followed by stimulation of terminals in the VPMpc or SI. No individual projection was sufficient to recapitulate the learning asymptote generated by stimulating all CGRP^PBN neurons; hence, some combination of projections relays salient aspects of the US, generating complementary signals that eventually reach the basolateral amygdala (BLA) to potentiate synapses receiving coincident CS information (Blair et al., 2001; Maren, 2001; Romanski et al., 1993). In support of this hypothesis, inhibiting CGRP^PBN-neuron terminals in either the VPMpc or SI during the US attenuated the association strength to the same degree as inhibiting the entire population, suggesting that preventing activation of either downstream partner impairs associative learning. We also observed that a substantial degree (~50%) of the conditioned response was maintained when inhibiting CGRP^PBN neurons, indicating that they are part of a distributed network that collectively relays the affective, motivational signal to forebrain neurons that form and store the associative memory (Lanuza et al., 2004; Lanuza et al., 2008; Shi and Davis, 1999). Interestingly, work examining the ability of the CGRP^PBN neuron projection to the VPMpc to generate an associative memory using taste as a CS indicated no conditioned taste aversion formation when paired with brief photostimulation, while projections to the CeA and ovBNST did (Chen et al., 2018), suggesting that either the relayed signal is the wrong modality for combining with CS taste information, or that a specific temporal activation pattern different from that tested is required to form an association (e.g. longer term activation that better mimics visceral illness). Hence, it is surprising that associative learning to a tone generated conditioned freezing behavior, while association with a taste did not alter preference, even though the VPMpc is an integral part of the ascending taste network (Liu and Fontanini, 2015). More work assessing the response patterns of individual VPMpc neurons to diverse sensory modalities and their contribution to conditioned taste aversion is required to resolve these paradoxes.

Previous work in rats indicates that PBN projections to the CeA are capable of eliciting both escape behaviors and fear learning in addition to driving place avoidance (Sato et al., 2015), while in mice it has been demonstrated that activation of CGRP-receptor neurons in the CeA is sufficient to act as a US to drive associative fear learning, and that silencing these neurons prior to conditioning attenuates conditioned-fear responses (Han et al., 2015). We activated both the rostral and caudal CeA terminal fields of CGRP^PBN neurons and were surprised that neither manipulation was individually capable of generating a fear memory, suggesting a number of possibilities that may give rise to these findings: (1) It is possible that CGRP^PBN neurons are only one component of the PBN projection to CeA and other PBN cell types are important for aversive learning, (2) CGRP-receptor neurons in the CeA may make up a larger population than those activated by CGRP^PBN-neuron terminal stimulation and receive input from secondary sources that are important for aversive learning; (3) direct activation of neurons in the CeA may be more efficient than terminal stimulation and may thus be able to drive associative learning; and (4) since the mice used in our study are heterozygous for CGRP, wholly intact neuropeptide signaling may be required for association formation derived from PBN to CeA stimulation (Shinohara et al., 2017; Okutsu et al., 2017). In support of the first two possibilities, we observed that inhibiting CGRP^PBN neuron terminals in the CeA during foot-shock delivery had no effect on associative fear learning, suggesting that the relayed activity from CGRP^PBN neurons to the CeA during the foot shock is not necessary for the CS-US association. This is in apparent contrast to previous work demonstrating that silencing CGRP-receptor neurons in the CeA prior to conditioning attenuates conditioned responses to the CS (Han et al., 2015). However, that manipulation was permanent and failed to distinguish between association formation and recall, which our US-only inhibition did, suggesting that reactivation of CGRP-receptor neurons in the CeA after conditioning may underlie the observed reductions in conditioned responding. Based on these observations and our data implicating CGRP^PBN neuron projections to the CeA in generating robust unconditioned-responses, we propose an alternate model, wherein CGRP^PBN neuron connections to the CeA, PSTN, SI, and ovBNST drive unconditioned responses to the US, post-conditioning activation of BLA neurons by the CS reactivates the CeA to generate conditioned responses (Kim et al., 2017), and CGRP^PBN neuron connections to the SI and VPMpc primarily mediate the role of CGRP^PBN neurons in associative-fear learning, a compelling arrangement given that in rats, these two downstream partners are the most directly invested in cortical circuits (Cechetto and Saper, 1987; Wenk, 1997). Our data establish partially separable ascending routes from CGRP^PBN neurons for generating unconditioned responses and forming associative memories to aversive stimuli.

Calca^Cre/+ mice (C57Bl/6 background) were generated and maintained as described (Carter et al., 2013). Male and female Calca^Cre/+ mice were used for all studies. Following stereotaxic surgery, mice were singly housed for at least three wk prior to and during experimentation with ad libitum access (unless noted otherwise) to standard chow diet (LabDiet 5053) in temperature- and humidity-controlled facilities with 12 hr light/dark cycles. All animal care and experimental procedures were approved by the Institutional Animal Care and Use Committee at the University of Washington.

AAV9-Flex-ChrimsonR:tdTomato was purchased from UNC GTC Vector Core (AV6556B; 4.5 × 10¹² viral particles/mL). AAV1-DIO-ChR2:YFP, AAV1-DIO-JAWS:GFP, rAAV2-retro-Flp, AAV1-Cre_on-Flp_off-ChR2-YFP, AAV1-Cre_on-Flp_on-ChR2-YFP and AAV1-DIO-YFP viral vectors were produced in-house by transfecting HEK cells with each of these plasmids plus pDG1 (AAV1 coat stereotype) helper plasmid; viruses were purified by sucrose and CsCl-gradient centrifugation steps, and re-suspended in 0.1 M phosphate-buffered saline (PBS) at about 10¹³ viral particles/ml.

Bilateral stereotaxic injections of virus (0.28 µl per side) into the PBN of Calca^{Cre /+} mice were achieved as described (Carter et al., 2013). In mice used for ChR2-optogenetic experiments, two custom-made fiber-optic cannulas were implanted bilaterally above the PBN (AP 4.70 mm, ML ±1.50 mm, DV 2.90 mm), VPMpc (AP 1.90 mm, ML ±1.25 mm, DV 3.65 mm), PSTN (AP −1.80 mm, ML ±1.50 mm, DV 4.60 mm), cCeA (AP 1.50 mm, ML ±3.10 mm, DV 4.30 mm), rCeA (AP 0.70 mm, ML ±2.85 mm, DV 4.50 mm), SI (AP 0.30 mm, ML ±1.80 mm, DV 4.40 mm), or BNST (AP +0.20 mm, ML ±1.20 mm, DV 4.00 mm). For three-fiber, dual-stimulation experiments, three custom-made fiber-optic cannulae were implanted in the left hemisphere, one above the rCeA/SI (AP 0.60 mm, ML – 2.50 mm, DV 4.40 mm), one above the cCeA (head inclined at a 10° angle; AP 2.15 mm, ML – 3.30 mm, DV 4.10 mm), and one above the VPMpc (AP 1.95 mm, ML – 1.00 mm, DV 3.80 mm). For JAWS-photoinhibition experiments, fiber placement was same for PBN, VPMpc and SI; fibers for CeA were placed at AP 1.10 mm, ML ±3.00 mm, DV 3.85 mm. For all experimental mice, fiber-optic cannulae were affixed to the skull with C and B Metabond (Parkell) and dental acrylic. Mice were allowed to recover for three wk before the start of behavioral tests. For collateralization-tracing experiments rAAV2-retro Flp virus was injected (0.48 µl unilaterally) into the VPMpc (AP 1.92 mm, ML ±1.00 mm, DV 3.85 mm), PSTN (AP −1.90 mm, ML ±1.50 mm, DV 4.70 mm), CeA (AP 1.10 mm, ML ±3.10 mm, DV 4.10 mm), or ovBNST (AP +0.20 mm, ML ±1.00 mm, DV 4.00 mm) and INTRSECT virus (0.35 µl unilaterally) was injected into the PBN. Tracing mice were sacrificed 4 weeks after virus injection.

Mice were excluded from individual test data if (1) they became immobilized due to tangled fiber-optic patch cords during the behavioral tests, (2) they escaped the arena during photostimulation, or (3) there was limited error-free data collected in pulse-oximeter physiological measurements (this only occurred with respiratory measures). Mice were excluded from all analysis if post-hoc histological examination revealed that viral expression was weak or unilateral, or that fiber-optic cannulae were not appropriately targeted over the projection-site of interest. Locations of fiber tips for all animals that passed the expression and placement criteria are summarized in Figure S3. There was also progressive dropout due to headcap loss requiring animal sacrifice during the study; all data were included up to that point pending histological analysis.

Mice were anesthetized with Euthasol (0.2 ml, i.p.) and intracardially perfused with 4–6°C cutting solution containing (in mM): 92 N-methyl-D-glucamine, 2.5 KCl, 1.25 NaH₂PO₄, 30 NaHCO₃, 20 HEPES, 25 D-glucose, two thiourea, 5 Na-ascorbate, 3 Na-pyruvate, 0.5 CaCl₂, 10 MgSO₄. Coronal slices (300 μm) were cut with a vibratome (Leica VT1200) and kept in the same cutting solution at 33°C for 12 min. Slices were transferred to a 25°C recovery solution containing (in mM): 124 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 24 NaHCO₃, 5 HEPES, 13 D-glucose, 2 CaCl₂, 2 MgSO₄. Recordings were made in artificial cerebral spinal fluid (aCSF) containing (in mM) 126 NaCl, 2.5 KCl, 1.2 NaH₂PO₄, 26 NaHCO₃, 11 D-glucose, 2.4 CaCl₂, 1.2 MgCl₂ continuously perfused at 33°C. All solutions were continuously bubbled with 95%:5% O₂:CO₂ (pH 7.3–7.4, 300–310 mOsm). Patch-clamp recordings were obtained with a MultiClamp 700B amplifier (Molecular Devices) and filtered at 2 kHz.

All data were analyzed using Prism 8.0 (GraphPad Software) as described in Supplemental Information. In brief, no tests were used to determine normality of data distributions or to pre-determine sample size; sample size was chosen based on past experience with expected effect sizes. Within-subject data was analyzed using two-sided, paired t-tests; across subject analysis was done with a combination of Welch’s t tests (unpaired, correction for no assumption of equal standard deviations), ordinary one-way ANOVA (with Tukey’s or Dunnettt’s correction for multiple comparisons), and ordinary or repeated measure two-way ANOVAs (with Sidak’s correction for multiple comparisons). For two-way ANOVAs, p-value for Treatment (i.e. ChR2 vs YFP)<0.05 is indicated to the right of each graph, and post-hoc row analyses’ p-values<0.05 are listed above individual data points.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

We thank G Stuber, L Zweifel, C Campos, A Dhaka, B Land, and J Kim for insightful discussions, B Land for guidance with analgesia assays,, J Ramirez for access to plethysmography chambers, G Stuber for use of electrophysiology rig, and J Allen for production of AAV viruses. AJB was supported by a National Institutes of Health T32 Graduate Training grant (T32NS099578). RDP received support from a National Institutes of Health grant (R01-DA24908).

Animal experimentation: This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All of the animals were handled according to approved institutional animal care and use committee (IACUC) protocols (#2183-02) of the University of Washington. All surgery was performed under isoflurane anesthesia, and every effort was made to minimize suffering.

1. Received: June 8, 2020
2. Accepted: August 28, 2020
3. Accepted Manuscript published: August 28, 2020 (version 1)
4. Version of Record published: September 10, 2020 (version 2)

© 2020, Bowen et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.